Get documents about "Natural Language Processing Arabic Cursive Handwriting Recognition
W
Shared by: uxx99201
Categories
Tags
handwriting recognition, character recognition, document analysis, cursive handwriting, arabic word, pattern recognition, cursive script, word recognition, international conference, handwritten word recognition, arabic script, pattern analysis, recognition systems, hidden markov model, arabic characters
-
Stats
- views:
- 59
- posted:
- 7/9/2010
- language:
- English
- pages:
- 93
Document Sample
Natural Language Processing:
Arabic Cursive Handwriting
Recognition
A. Belaïd
A. Belaïd
1
Preamble
Arabic
– Part of the Semitic language family
One of the most spoken language in the world
– Nearly 250 million people speak Arabic
Spoken outside Arabic countries
– Over 600 000 people In the United States speak
Arabic
Give rise to other alphabets
– Farsi, Urdu…
• spoken by millions of people from Iran, Pakistan,
India…
2
Preamble
The work presented here is that of numerous
researchers working with me…
– Najoua Ben Amara
– Samia Maddouri
– Afef Kacem
– Imen Ben Cheikh
– Hiba Khelil
– Mohamed Yazid Boudaren
– Nazih Ouwayed
– Christophe Choisy
– Umapada Pal
3
Presentation outline
Introduction
– Brief historic, specific applications
Writing characteristics
– Those shared with Latin
– Proper to Arabic
Issues of cursive word recognition
– Reading models
– Global word-based (holistic), Local letter-based
(analytical) approach, Hybrid approach
Language Processing
– Some basic solutions for handwriting recognition
4
Introduction
The field arose before the apparition of
computers
Maturity
Small devices
Price Intelligent pen
Handwritten
forms
1st Postal
address reader
Forms
OCR in
industry
Working
Models
Patents on
OCR: blind
telegraph
1900 1916 1950 1965 1980 2008 5
Today for Latin:
An industry and a real market
Material :
– Scanners adapted to documents
We know
– scan documents in a huge quantity,
preserving the image quality
– compress them, publish them on the net
– recognize by OCR and identify some
structure elements
But …
– on very good quality documents
– rather recent, poor structure, printed,
– Handwriting:
• Just few work available for specific
applications: small vocabulary
6
Introduction
Arabic
Started at 1980
– increasing demand for information indexing and retrieval
Pioneers
– A. Amin (Loria),
– M. Cheriet (ETS, Montreal),
– N. Ben Amara (ENIT)…
Today
– Many Labs: REGIM (Sfax), LRI (Annaba), READ (Loria),
ETS (Montreal), CEDAR (USA), ISI (India)
Many dedicated sessions and workshops
– IWFHR, ICFHR, CIFED, ICDAR, SACH’06
Several public datasets
– IFN/ENIT, DARPA/SAIC, CENPARMI, Farsi-City…
• See Volker Märgner list
7
Introduction
Arabic
Commercial Arabic OCR
– Number of commercial Arabic OCR engines
• Sakhr's Automatic Reader: ~1500$
• Readiris from IRIS : ~500$
• Verus from NovoDynamics : ~1300$
• Omnipage
– Evaluation by UNLV
• Sakhr (90,33%), OmniPage (86.89%)
Open Source Arabic OCR projects
– The Siragi project (started in 2005)
• Part of the Arabic Unix open source project
8
Introduction
Some specific applications
Bank check recognition of courtesy amounts
9
Introduction
Signature recognition,
verification, forgery detection…
V. K. Madasu et al. M.A. Ismail et al.
Pattern Recognition, 38 (2005) Pattern Recognition, 33 (2000)
Normalized vector angle (α) in boxes Global and local features in boxes
Algorithms based on fuzzy concepts 10
Introduction
Some specific applications
Writer identification
Automatic Writer Identification Using Connected-Component Comparison of Gabor-Based Features for Writer
Contours and Edge-Based Features of Uppercase Western Identification of Farsi/Arabic Handwriting,
Script, L. Schomaker et al, PAMI, V. 26, N. 6, 2004 IWFHR,’06, F. Shahabi et al.
11
Introduction
Word Spotting by request
Candidates Prototypes
Stochastic model: MADAME Template matching: ALMALIK
Ch. Choisy, A. Belaid, Cross-learning in Analytic Spotting Words in Handwritten Arabic
Word Recognition Without Segmentation. IJDAR’02. Documents, S. Srihari et al. , SACH 2006 12
Introduction
Newspapers segmentation
Arabic Page
Segmentation,
Planet, K. Hadjar et
al. ICDAR03
Connected Pattern Segmentation and
Title Grouping in Newspaper Images, P. E 13
Mitchella et al, ICPR’04
Introduction
Some specific applications
Paleographic inspection
Progressive evolution
between VI and XVIc
University of Pisa: to classify and
identify medieval scripts
University of Annaba:
INSA Lyon: Auto-similarité de formes pour la
discrimination des styles d’écriture des
manuscrits médiévaux, I. Moalla, F. LeBourgois,
14
H. Emptoz, A. M. Alimi
Introduction
The issue of recognition
Handwritten Latin recognition showed first the way
– In terms of modalities
• On-line vs Off-line
– In terms of scripts
• Printed vs Handwritten
– In terms of pre-processing
• Shape normalization
• Feature extraction: indices or graphemes
– In terms of methodologies, classified regarding:
• use or not lexicon
• nature of primitives / model: structural, statistic,
stochastic
• vision level: local or global
15
Introduction
Initiated by Speech recognition
Today: a well established PR System
Feature extraction
Preprocessing &
Vector quantization
ﻧﻘﺶ
« Bonsoir, à demain Recognition Sequence
start ﺑﻘﺔ
pour une nouvelle
stand ﻟﺜﺔ
édition du journal »
ﺗﻔﺖ
store Enrollment
ﺑﻘﺔ
Tree structured HMM Models
lexicon database
16
Introduction
When LP contributes?
Text
Phoneme /
Training Preprocessing + Language
Character /
data Feature extraction Modeling
Modeling
Enrollment Lexicon +
Character Models Grammar
Recognition
Image Preprocessing + Word
Recognition search
input Feature extraction Sequence
17
Introduction
The process bases are well
established
Discriminate Model
Global
a b ... z Learning Recognition
Segmentation
Analytic
Discriminate Path
Pre-segmentation Internal Sliding Window
18
Introduction
Performances: criteria influencing
the quality
Writer nb
omni
multi
Lexicon
mono size
guided reduced large
non constrained
Writing
19
Introduction
The performances are satisfactory
for Arabic
Script Process Model
20
Outline
1. Introduction
2. Writing characteristics
3. Issues of Segmentation
4. Natural Language Processing
21
Writing characteristics
Some of them are similar to Latin
Writing lines
Upper-line
Mean-line
Base-line Alignement X-height
Lower-line
Baseline
22
Writing characteristics
Some of them are similar to Latin
Perceptive invariants: J. C. Simon called: regularities
and singularities
Letter peculiarities
Letter support
23
Writing characteristics
Some of them are similar to Latin
Perceptive invariants / regularities and singularities
24
Writing characteristics
Some of them are similar to Latin
Perceptive invariants / regularities and singularities
25
Writing characteristics
Some of them are similar to Latin
Perceptive invariants / regularities and singularities
26
Latin vs Arabic
What changes?
Essentially the script, always complex
Latin Systems Arabic Systems
The difficulty is gradual The difficulty is permanent:
Cursiveness, ligature, tashkeel
27
Latin vs Arabic
What changes?
The gaps are significant for Latin, not always for Arabic
Latin Arabic
Between words Everywhere
28
Latin vs Arabic
What changes?
The ligatures are permanent: horizontal and
vertical
29
Latin vs Arabic
Arabic has some peculiarities
1. Helpful: accents and diacritical dots contribute to
the recognition
30
Latin vs Arabic
Arabic has some peculiarities
2. Helpful: the letter elongation contributes to the
segmentation
31
Latin vs Arabic
Arabic has some peculiarities
3. Helpful: the position of the hamza (16) and
the descenders
32
Latin vs Arabic
Arabic Pecularities?
5. Helpful: PAWs offer a pause in the writing, a
decomposition of the writing
• Simplify the script apprehension, make easier the
linear recognition
PAW
[Al-Badr and R. M. Haralick 1998]
33
Arabic Script
In conclusion
• Arabic: more global than syllabic
• PAWs : facilitate the recognition
• PAW level ~ letter level in Latin
• For recognition
– In Arabic: to reach PAW level: characteristic
information
– In Latin: to reach letter level
• The PAW level is the stable level
makes it semi-global
34
Outline
1. Introduction
2. Writing characteristics
3. Issues of Recognition
4. Natural Language Processing
35
Issue of segmentation
Due to the local variability
– It is widely accepted
• Arabic word segmentation in letters is very
delicate and not always ensured
• Usually in most attempts, Arabic word is
segmented into graphemes (copied on Latin)
– This is an error!
36
Issues of recognition
Considering Arabic peculiarities
Reading models
– The recognition of a word
• implies the processing of visual data and its
interpretation at the linguistic level
– Psychologists call "mental lexicon access"
• the process by which the human associates the image
of a word to its significance
Several models emerges
37
Interactive Activation Model
Mc Clelland and Rumelhart 1981
Important assumptions Words
– Perception takes place in a
multilevel processing: -
MATE MOVE
• Feature, letter, word
A consequence:
+ - + - - -
– more abstract levels of
representation are only Letters
accessed via intermediate E - M - O
level
A third assumption
– Processing combines both - + ++ - +
bottom-up and top-down
information refers Features
• readers can use their
(top-down) knowledge of \ /
words to help identify
letter sequences from
(bottom-up) visual input Neurons have excitatory and
inhibitory connections 38
IA & Arabic Recognition
Arabic writing
– fits very well the reading principle of IA
• Clearly privileges the superiority of the whole
• Local perceptual information is just used to help
word understanding
But the corresponding model
– should be adapted to consider the PAW level and letter
distortions:
• PAWs introduce an intermediate global level
Hence, perfect similarity if adapted
39
Perceptro [Côté, Cheriet 98]
Limited number of features
– Ascender, descender, loop
word not having these
features cannot be initialized
No training
no inhibition rapid
saturation
Recognition
– Perceptive cycles
– Top-down & bottom up
40
Transparent Neural
Network
[Maddouri, Belaïd, Ellouze, 03]
– Input correction by FD
[Ben Cheikh, Kacem, 07]
– Slight extended vocabulary
(Tunisian city names)
– Training possibility
[Ben Cheikh, Belaïd, Kacem, 08]
– Wide vocabulary
41
Arabic Recognition
Correction process
Propagation
ﻟﺺ
ﻟﺤﻢ
Back-Propagation
ﻟﺴﺮ
?
Original image
Reconstructed
image (harmonics)
Real image
42
Arabic Recognition
Experiments
– 2100 images, 70 words, 63 PAWs
– Without Perceptive cycle
• PAW RR: 68.42%
• Word RR: 90%
– With perceptive cycles
• PAW RR: 95%
• Word RR: 97%
43
Arabic Recognition
Methodologies
Considering human perception of Arabic writing
with the particularity of PAWs
revised literature approaches: vision degree
• Global-based vision classifiers
• Semi-global-based vision classifiers
• Local-based vision classifiers
• Hybrid-level classifiers
and examined their proximity with IA
44
Methodologies
I Global-based Vision Classifier
The word
– regarded as a whole
The features
– doesn’t need to be precise:
• presence and some
relationships
The approach
– assimilated to segmentation free
even if a segmentation is used, no local
interpretation is made
information is gathered at the word level
Its use is limited to small vocabularies
45
GBVC
Examples
Srihari and al [2005]:
– Several preprocessing steps
– Feature extraction for PAWs
and Words:
– aspects measurements
– Word resemblance by NN
Noise suppression and binarization
• 10 writers writing 10
documents each : word
extraction is ~ 60%,
rr=70%
Suppression of internal contours
Fusion of minor components
46
GBVC
Examples
Al Badr et al [1998]
– Free segmentation method :
• detects a set of shape
primitives on the word
• matches the regions of the
word with a set of symbol
models
• maximizes the a posteriori
probability of the arrangement
Correspondence regions
of symbol models of the model (in shades of gray)
– Word recognition scores :
clean (99.39%), degraded
(95.60%) or scanned
(73.13%)
Matching with symbol model
47
Local-Based-Vision Classifier
Example
Shirin Saleem et al. [2008]:
– BBN Technologies, Cambridge, MA: BBN Byblos OCR
System (DARPA data set):
– Locate line tops and bottoms
– Extract narrow overlapping vertical slices of the image
• measure features on each slice
• reduce the size of feature using Linear Discriminant
Analysis (typically 15 features)
48
Simple Frame-based
Features
Examples of features:
– Intensity as a function of vertical position
– Vertical derivative of intensity
– Horizontal derivative of Intensity
– Local angle within a small window
– Difference of angle
49
Character Hidden
Markov Model
50
Local-Based-Vision Classifier
Example
R. Al Hadj et al [2006]
– HMM for letters and words
with sliding windows
– Windows correspond to 3
different orientations: density
description
– A second system integrates
all the orientations in each
position
85.02% (Top1) 91.29% (Top2) 93.14% (Top3)
51
Global-Based Vision Classifier
Examples
Khorsheed et al [2000]
– Polar transformation coupled
with a Fourier transform
– Each word: template with Fourier Original images
coefficients
– Recognition
• normalized ED from templates
• In a multi-font approach: Normalized images
– 95.4% of good word by polar transform
classification on 1700
samples of different size,
angle and translation
52
GBVC
Synthesis
The works related
– accredit the word superiority Word
– Many feature combinations
and models perform well
The proximity with IA?
Feature
– can operate
– but limited to 2 levels
needs more precision
in feature extraction
Adaptation of GC to Arabic
– Possible if high level features Input
used
53
Methodologies
II Semi Global Vision Classifier
The word
– natural concatenation of independent PAWs which
provides a natural segmentation
Important to find features
The features
– are numerous and different
require normalization of image before extraction
The approach
leads to reduce the vocabulary as only the PAWs are
considered
54
Semi-Global-Based VC
Examples
Planar HMM: Ben Amara, Belaïd, Ellouze [1996]
For the main: band width:
• observation P of the S HMM
• a specific function (normal
density) of the duration
For the secondary:
band description: List
of B&W segments in
– Morphology of each PAW
each line of the band
– 99.84% for 33168 samples, 100 PAWs
55
Semi-Global-Based VC
Examples: town name recognition
Burrow [2000]
– Method
• to trace lines making up the
town name, and to use these
as a representation
– Features
• Vector angles + average
length +…
– Results
• ED (converted into pseudo
probabilities) between the
test feature vectors and all
those in the training set
• Recognition rate 74%
56
SGBVC
Synthesis
The works related
– similar to those for GBVC
– some are reported on PAWs
The proximity with IA is limited
– only features and PAW levels considered
The adaptation of Semi Global BVC to Arabic
fits well but limited to PAWs
fits better if a gathering procedure of PAWs is possible
57
IA architecture for Semi-
GlobalVC
…
PAW
… Letter
Feature
Input
58
Methodologies
III Local-Based Vision Classifier
The word
– regarded as a list of letters or
smaller entities
The features
– should be located precisely,
inversely to the other
approaches
where flexibility is tolerated
The approach
– should gather, confront these
entities to identify the word
The interest
– can cope with large vocabulary
59
LBVC
Example
Multi-level handwritten word recognition for tunisian city
names Miled [1997]
60
LBVC
The strategy: 2 perceptive levels
First perceptive level:
– practices the global view by extracting visual indices: by
tracing and grapheme extraction
– This global Information is extracted in the main zones :
(b) diacritics; (c) baseline and middle zone characters
61
LBVC
Example
Then, visual indices are extracted by tracing
62
LBVC
Example
Finally: a Markovian modeling is operated on the list of visual
indices
Recognition: 58,9% (top1) to 86,8% (top10)
63
LBVC
Example
The 2nd perceptive level practices an analytical approach by
extracting finest features: graphemes
18 classes
• 1. A: alef, 2. B - D: graphemes with ascenders
• 3. E – H : graphemes with both ascenders and descenders
• 4. I – M : graphemes with descenders
• 5. N – R : graphemes within the middle zone
Recognition: 69.68% (top1) to 91.66% (top10)
64
Local-Based-Vision Classifier
Example
Finally, the 3rd level practices a pseudo-analytical
modeling and recognition of PAWs and words
37 words: 80.11% (Top1), 90.79% (Top5) 65
LBVC
Synthesis
The works related
– give good result showing that the analytic approach can
perform well
– point out drawbacks of over and under-segmentation
As letters or segments are recognized independently
any error can perturb the whole recognition process
The proximity with IA is far
- WSE is not taken into account because
- no global vision of the word, but as a sum of small parts
66
Methodologies
IV Hybrid Level Classifier
The word
– regarded as a whole as well globally as in details
The features
– Correspond to precise location reinforced according to the level
of detail needed
The approach
– combines different strategies: to approach more human
reading:
• the analysis must be global for a good synthesis of
the information
• while being based on local information suitable to
make emerge this information
67
Hybrid-Level Classifier
Examples
NSHP-HMM [Choisy & Belaïd 02] :
– a random field drawing its observation directly in the
image
– a HMM taking into account the column observations
X θ ij
X ij
68
Hybrid-Level Classifier
Examples
Analytical aspect : Local-Global aspect :
69
Hybrid-Level Classifier
Examples
NSHP-HMM [Vajda & Belaïd 06] :
– Combination of structural and pixel information
70
HM &
Synthesis
The works related
– seem efficient
– IA seen as meta-model reassembling models working at
different visual levels: global, local, semi-global
The proximity with IA is close
– If we add the PAW level
- It combines different levels as proposed in IA
The interest
- to do the maximum without segmentation
- if needed, we can operate a segmentation which will be
guided by the context
71
Outline
1. Introduction
2. Writing characteristics
3. Issues of Recognition
4. Language Processing
72
NLP
Number of effective Arabic words go past 60 billions!
– due to its morphological complexity [K. Darwish 02]
makes their automatic processing unrealistic
– handicaps: dictionary building, IR, automatic spelling…
Simplification of their pattern becomes mandatory
for their processing
One solution seems to turn towards
morphological analysis and word stemming
73
NLP
Many studies
– highlight the richness and the stability of Arabic in
terms of morpho-phonologic peculiar to this language
[A. Ben Hamadou 93], [S. Kanoun 02], [W. Kammoun
04], [M. Cheriet 06]
Questions
– Importance of the kind of linguistic knowledge
– more appropriate location for its incorporation
74
NLP
Most of them confirm
– The morphological structure of Arabic
• can be analyzed in terms of consonantal roots,
considered as independent morphological unit
Tri-literal roots, the most common of them
– [Watson 06]
• give rise up to 15 verbal forms or stems, one basic
and the rest derived
– [Ben Hamadou 93]
• an average of 80 currently used words derive from a
given root
– [Kanoun 02]
• 808 healthy tri-consonant a lexicon of 98 413 words
75
Word decomposition
An Arabic word is
decomposable (e.g. derivates from a root) ( :ﻣﺪرﺳﺔschool) or not
( :دڪﺘﻮرdoctor)
A decomposable word is composed of
morphemes: prefix, radical and suffix
The radical (or the verbal core) is Radical
– the derivation of a root according
to a given scheme by introducing
“access” letters: م ,ا
A root is either
– tri-consonant (three letters): آﺘﺐ
– quadri-consonant (four letters): دﺣﺮج
– Healthy ( )ﺟﺮحor non-healthy ( ﻗﺎلcontains a vowel at least)
76
Arabic Morpho-phonological
Concepts
Schemes can go up to 70
– ﻤﻨﻔﻌﻞ, اﻔﺘﻌاﻞ, ﻣﻔﻌﻮل, اﺳﺘﻔﻌال, ﻤﻔاﻋﻞ, ﻤﻔﻌاﻞ ,ﺘﻔاﻋﻞ ,ﻔﻌل
Schemes classes are:
– Verb “”رﺣﻞ“ / ”آﺘﺐ
– Agent noun “”راﺣﻞ“ / ”آﺎﺗﺐ
– Accentuated agent noun ﺣ
“”ر ّﺎل
– Patient noun “”ﻣﻜﺘﻮب
– Machine noun “”آﺘﺎب
77
The approach:
Transparent Neural Network
Easy to train:
– Decomposable on 3
mono-layers
– Training is rapid
But not allows too many
outputs
78
To process a wide vocabulary:
several improvements
First craftiness
– To consider the word as the conjugation of a root
according to a given scheme
– To separate the outputs in roots and schemes
• For 8000 words that rise from 100 roots the
maximum of schemes is 1400
100 roots
word TNN
RNT
1400 conjugated schemes
8000 size problem 1500 size problem (still high !)
79
Second craftiness
To consider the scheme as a brief scheme ( ﻳﺘﻔاﻋﻟﻮﻦis defined by a
brief scheme: a non-conjugated one, )ﺘﻔاﻋﻞand a set of conjugation
elements
– Brief schemes number is around 75
– Conjugation elements number is 12 (tense, gender, person, definition)
87 neurons represent 1400 conjugated schemes
8000 1500 187 (100 roots + 75 schemes + 12 conjugation elts.)
80
Third craftiness
Roots and schemes trainings are independent
– These two trainings do not require the same information
• The Information about word PAWs are:
– useless for the training of its root
آﺎﻓﺢ
root: آﻔﺢ
آﻔﺎح
– useful for the training of its scheme
آﺎﻓﺢ scheme: ﻓﺎﻋﻞ
آﻔﺎح scheme: ﻓﻌﺎل
81
Third craftiness
To separate them and so lighten them by
splitting the TNN into two models:
TNN_R 100 outputs
Word
TNN_S 87 outputs
These sizes are now practicable
82
Neuro-Linguistic
approach
TNN_R: three-layer network
– Learns how to focus on root letters and ignores access ones
reserved for schemes, trains roots from structural primitives
Letters(117)
Primitives (70)
Roots(100)
2. 64 ﺗﺎ
PD
-1
2. 63 ـﺒﺎ
QM
8. 94
0.74
RD RF HF QM PD HF
4. 94
آﺎ ﺑﻌﺪ
2. 8
RD
2. 64 ﻋﺎ
5. 9 -0. 46
RF ﺳﺎ
4. 93
15. 3
د root: ﺑﻌﺪ 83
Neuro-Linguistic
approach
TNN_S: 4 layers
– learns schemes from structural primitives, how to ignore
root letters, focuses on access letters: prefix, suffix …
– PAWs of Arabic schemes more reduction
PAWS of Schemes
Letters Schemes
word: ﺗڪﺎﺛﺮ ﺗﺎ*ﺗﺎ
Primitives ﺗﺎ ﺗﻔﺎﻋﻞ
PD ﻣﺘﺎ*ﺗﺎ
آﺎ
+ PD+HM+HF+JF ﻣﺘﻔﺎﻋﻞ
+ Access letters: ت,ا
HM
آﺎ Conjugation elements (10)
+ Singul + accompl.+ masc. HF Singular
ﺛﺎ
JF ** masculine
ر
feminine
Accomplished
84
Neuro-Linguistic
approach
TNN_R training: words containing the root
– As the information trained corresponds to two independent cases:
• letter constitution / root formation: letter location and sisters
• and there is no local error to treat
we separate it into two MonoLP: 1 & 2
اﻧﺼﺮف
أﺗﺼﺮف ﺗﺼﺮف &
ﻳﺘﺼﺮف
Structural
features
Letter position &
85
sister letters
Neuro-Linguistic
approach
TNN_S training: same corpus as for TNN_R
The same way: 3 sub-networks:
Mono-LP1 Mono-LP3 Mono-LP4
اﻧﺼﺮف
ﺗﺼﺮف أﺗﺼﺮف & &
ﻳﺘﺼﺮف
Fixed manually to
Letter position & indicate those should
sisters be activated 86
Recognition: perceptive cycles + linguistic restrictif
Perceptive cycles
!!
X
ﺳﺮق
ﺻﺮف اﻧﺼﺮف : HI PD BM JrF BPI
TNN_R ﺣﺮف X
اﻧﺤﺮف : HI PD RM JrF BPI
X
ﺻﺮخ
HI PD BM JrF BPI
!!
اﻧﺼﺮف
Linguistic
restriction
HI PD BM JrF BPI
اﻧﻔﻌﻞ
TNN_S !!
X
اﻓﺘﻌﻞ
Perceptive cycles
Experiments
Vocabulary size
– 1531 words
– 51 roots
– 25 brief schemes
Training base
– The same training corpus for TNN_R and TNN_S
• TNN_R corpus size : 1531 (words) to train 50 roots
• TNN_S corpus size : 1531 (words) to train 25
schemes
Test base
– size: 765= 255 (words) * 3 (samples)
88
Word base
Comparison with approaches
dedicated to wide vocabularies
Vocab
Writing Approach . size Top1 Top2 Top3 Top4
[Kanoun 02] Typesetted Analytic 545 69 % 84 % 95 % 97 %
[Kammoun Typesetted Analytic 1423 81.3% 95.7% 96.4% 99.7%
06]
[Touj & Ben Handwritten Analytic 25 88,7%
Amara 07]
[Ben Cheikh Typesetted/ Pseudo- 80,7% 89.4% 91.9% 92% TNN_R
and al 08] Handwritten global 1531
93% TNN_S
Conclusion (1)
Conclusion
Neural model + linguistic knowledge
– Arabic writing recognition with wide vocabulary:
– Knowledge: Arabic morphology analysis
Favors the recognition of words which have
never been learned
– It is just needed that its root and its scheme have been
already learned via other words
90
Example
The words « »آﺜﺮة« ,»آﺜﻴﺮ« ,»أآﺜﺮand « »آﺜﺮتparticipate to the
training of the root «( »آﺜﺮin addition to their schemes)
The words « »ﺗﺪاﺧﻞ« ,»ﺗﻘﺎرب« ,»ﺗﻌﺎﻧﻖand « »ﺗﻤﺎﺳﻚparticipate to the
training of the scheme «( »ﺗﻔﺎﻋﻞin addition to their roots)
.
Hence, when recognizing the word « ,»ﺗﻜﺎﺛﺮour model should
be able to recognize :
the root «»آﺜﺮ && the scheme «»ﺗﻔﺎﻋﻞ
91
Conclusion (1)
Perspectives
The improvements will continue:
– Knowledge
• Considering other aspects of the Arabic
morphology: other kinds of roots, derivations…
– Recognition stage
• More linguistic restriction in the perceptive cycles
– Data Base
• To work on more realistic vocabulary by enlarging
more the size
92
Conclusion (1)
Thank you
Related docs
